Value
AI got cheaper. The bill didn't.
The price per use is falling fast. The spend is rising faster. Price the work, not the token.

Rujuta Singh
3 min read
Same AI performance, about 13 times cheaper every year. And still, 93% of the companies McKinsey asked went over their AI budget.
It's a usage problem. Not a pricing problem.
Two curves
What you pay per use: down. How much you use it: up. Faster.
This year I built a whole website with AI. Under 10 hours. $15 total. I thought the budget question was dead. For one person, it is. For a company, I was half right.
The price did drop. Then everyone used it for everything. The most expensive AI for a two-line email. An AI bot running all night on a ten-minute job.
So the price fell and the bill rose.
McKinsey found AI spend grows nearly four times when a company goes from pilots to everyone using it. Only 20 to 25% of companies have mature ways to track that cost.
This is not a slow-down take. Keep building. Value leaks every month you wait. Just stop pricing per AI use. Price the work.
- What is this task worth?
- Does it need the most expensive tool?
- Who owns the bill?
I haven't figured this one out yet. I'm figuring it out as we go.